Shishir Mehrotra: preference

Transcript context

…I didn't know this would be a very interesting game, but we could try to create one. Okay, so I'll describe what eigenquestions are, but maybe I'll start by telling you a little bit about where the concept came from. And, this is actually a YouTube-ism that... Maybe just to place ourselves in history, so in 2008 I joined YouTube. And many people don't remember this, but YouTube at the time was seen as a mistake. It was seen as Google's first bad acquisition, everything else had worked, but this thing kind of seemed like a disaster. It was grainy videos, we were losing lots of money, we had billions of dollars in lawsuits, none of it really seemed that obvious. There's lots of discussion on how to reorient it, fix it, and so on. And so everybody outside, that's what they saw. If you stepped inside a YouTube staff meeting in 2008, actually one of the toughest questions we were answering was this question we called the Modern Family question. And it may sound small, but it actually was very perplexing. And the question was pretty simple, the question was if you looked at our search traffic at YouTube, one of the top five queries basically every week was for a show called Modern Family. And Modern Family was number one show on television at the time, by far the most popular. And we were second-biggest search engine in the world behind our parent company Google, so search traffic was very important. There was one big problem, people would come search for Modern Family, one big problem, we didn't have Modern Family on YouTube. And so we'd give them some pretty crappy responses. And the question was... And by the way, ABC.com had decided to post every episode of Modern Family live on their website. Which nowadays is kind of typical, but in 2008 that was not typical, no is it kind of a big gamble they made, and they're going to post all these shows. So the question was, should we answer the query Modern Family by linking off to ABC.com? Do we link out or not? And the company basically divided. And so there's half the company, mostly the product team, vantage team and so on kind of aligned around a viewpoint that was, "That's what the user wants, link them off to ABC.com. We're not owned by Google, Google tries hard to prioritize, do right by the user, the rest will follow. That seems like the right thing to do, so let's do that." And then the other half of the company, most of the business function, sales, marketing, especially content partnership said, "Please, please, please don't do that. If you do that and you start linking off to all these other places, nobody's ever going to put good content on YouTube, and we're just going to get the stuff that doesn't deserve to live anywhere else. And that's not a very good path to be." And you can imagine that those two mindsets, it's almost like it was good versus evil debate, which is do right by the user or the business. These are all almost impossible to solve problems. to be." And you can imagine that those two mindsets, it's almost like it was good versus evil debate, which is do right by the user or the business. These are all almost impossible to solve problems. And this would happen meeting after meeting after meeting, did we make a decision yet? Modern Family, what are we going to do? So we had this offsite, we said we're going to spend the whole day, and we're going to figure this problem. And we went up to this hotel in Half Moon Bay, and the executive team all sat down, and I was asked to frame the discussion, go collect everybody's opinions, and collect all the data, and just ground it all in facts. And then we're going to have a discussion, we're going to reach a decision. And so the night before I'm sitting and thinking about, how the hell we going to have this discussion not just be a shouting match of this good versus evil position? And I happened to read a analysis that was being done by a different team at Google, the Google shopping team, where they were facing this interesting challenge of they were in this deep fight with Amazon, and they were getting their butts kicked, and they were trying to figure out why. And the walking in theory was, the Google shopping team's view was why would anybody ever go in to Amazon? You could come to Google, and we had indexed all of Amazon and the entire internet, why would you ever pick Amazon? And the feedback that was coming back from users was they'd say, "I picked Amazon because I value consistency over comprehensiveness. " They would say things like, "I really value that I go to amazon.com and I understand how the reviews work and how the ratings work, and I know that the returns work the way I want and I understand the shipping is going to work. And it just felt consistent. I know it doesn't have everything, but it has enough. And I value consistency over comprehensiveness." And so the night before this YouTube meeting, I decided to reframe the question and say, let's not have the discussion about linking out at all. We're going to start by having a theoretical discussion about, in a decade from now, is the online video market more likely to be about consistency or about comprehensiveness? And that is a question that you can have a very reasonable debater. What are the reasons why a market evolves towards consistency over comprehensiveness? And we basically have this discussion and we all came to the conclusion this market is going to value consistency over comprehensiveness. And by the way, I think, I mean now almost 15 years later, I think we're right. If you go look at video market obviously it exploded. There's so many great video properties out there, none of them are comprehensive. There is no one-stop shop or a place where all the video exists. And so I think we were right. But what happened was by answering that question, the link out question all of a sudden became super easy. We value consistency over comprehensive. We definitely don't need to link out. In fact, we should make a whole bunch of other decisions as well. he link out question all of a sudden became super easy. We value consistency over comprehensive. We definitely don't need to link out. In fact, we should make a whole bunch of other decisions as well. So we went and at the time we used to, this was the days of flashed players and so on, we embedded other people's players on YouTube. We stopped doing that as well. Probably the most famous decision we made was with the iPhone. As I mentioned earlier, the iPhone when it first shipped had no app store. And so they built all the first few apps including YouTube. And so here we were a few years later, iPhone the most popular phone on the planet and the YouTube app on the iPhone was built by a team at Apple and they had not been able to keep up with what we were doing. Almost half the catalog didn't play back on the iPhone, they were missing a bunch of features and so on. So I drove down to Cupertino and I sat down with Scott and Phil and said, "Hey, we're going to have to take back the YouTube app." And they said, "I don't understand. Why would you do that? You have default distribution on, as far as they were concerned, the most important operating system in the world. Why would you do that? You're going to have to rebuild all this from scratch and it just seems like a really bad choice." And I said, "No, no, it's actually quite an easy choice. We value consistency over comprehensiveness. We would much rather be on fewer phones with a more consistent experience than be on all of them with an inconsistent experience." So what, this is a choice we're making and it worked out fairly well. So this decision, the Modern Family question ended up becoming named as the example for a term called eigenquestion. So eigenquestions, it's not a German game show, it is a made up word and it's named after a math concept called eigenvectors. And the math is not really necessary, but for people who are curious, go back to linear algebra, eigenvectors are in a multidimensional space. They're the most discriminating vectors of the vectors in that space, the dimensions of that space, it's a concept that gets used a lot in machine learning and so on, but actually the math doesn't really matter. The eigenquestion, the simplest definition of eigenquestion, it's the question that when answered also answers the most subsequent questions. And it's a very simple idea that when you sit down and you say, hey, here's all these questions we ought to answer, how do we all usually rank them? Sometimes we rank them just by what order we came up with them. Sometimes we rank them by importance, which is the most impactful decision we're going to make. But this methodology says don't rank them that way. Rank them, like you said, [inaudible 00:53:53] thing about about leverage, same idea, rank them by which ones would eliminate the most other questions of the list. So you take that list and you said, should we link out to Modern Family? Should we own the YouTube iPhone app? Should we do... mind you, those were really hard questions to answer. But turns out if you answered just one question, do we value consistency over comprehensiveness, you answer all the others.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.